Imagine a cybersecurity landscape where artificial intelligence doesn't just react to threats—it anticipates them, learning from decades of digital dramas to keep systems safe and sound. That's the thrilling vision behind Cisco's groundbreaking new AI model, poised to revolutionize how we defend against cyber attacks. But here's where it gets controversial: is relying on AI for security a game-changer or a risky gamble? Stick around to explore this cutting-edge development and decide for yourself.
In an exclusive scoop, Cisco is ramping up its AI efforts with a fresh security model that packs more than double the parameters of its current flagship, Foundation-Sec-8B. For beginners, think of parameters as the building blocks of an AI model—more of them mean the system can learn from a wider array of data, potentially leading to smarter decisions. The company's existing hero, Foundation-Sec-8B (available on Hugging Face at https://huggingface.co/collections/fdtn-ai/foundation-sec-8b), boasts eight billion such parameters and is already powering several products. It's designed for practical tasks like automating the handling of security alerts, scanning code to spot vulnerabilities, and proposing workflows that put security front and center.
During the Asia-Pacific leg of Cisco Live in Melbourne, Raj Chopra, the senior vice president and chief product officer for security, unveiled plans for this upgraded model, which will boast 17 billion parameters. Just like its predecessor, it'll be trained on a rich tapestry of data: threat intelligence, summaries of past incidents, red team playbooks (those are simulated attack strategies used to test defenses), and other collected insights. Chopra emphasized that this new version will draw from Cisco's Talos threat intelligence team, which brings a treasure trove of 30 years' worth of data to the table. This long history of cyber insights could make the model extraordinarily adept at recognizing patterns in attacks.
To put this in perspective, consider how a seasoned detective uses years of case files to predict criminal behavior—here, the AI is doing something similar, but on a massive scale. And this is the part most people miss: the model isn't just about spotting threats; it's engineered to suggest actionable steps to mitigate them, blending detection with response in a seamless way.
Chopra clarified that this isn't a direct replacement for Foundation-Sec-8B. Instead, it's a bold new initiative tailored for that dual purpose of identification and remediation. 'And for that, we need an expanded model,' he told The Register, hinting at its launch right after Christmas—though not too long after, to keep the momentum going. He also teased a broader 'phalanx' of upcoming AI models and projects, though specifics were off the table for now. As an example of their ongoing work, Chopra highlighted the late-October update to SecureBERT (detailed at https://blogs.cisco.com/ai/securebert-2-0-ciscos-next-gen-ai-model-powering-cybersecurity-applications), a tool that assists security pros. Cisco's enhancements reportedly boosted its performance significantly, making it a more reliable ally in the fight against cyber foes.
But here's where controversy brews: Cisco develops these large language models (LLMs) because it believes businesses require a blend of universal security knowledge and customized data about their own operations to harness AI effectively in defenses. The models are open source, fostering collaboration, yet not without a business angle—Cisco integrates them into its own products and even recommends its Splunk tools as the go-to for customers to crunch their unique data. Is this open-source generosity genuine innovation, or a clever way to lock in customers? It's a debate worth having.
- For more on Cisco's consolidation of management tools, check out: https://www.theregister.com/2025/06/10/ciscolivecloudcontrolnews/
- Curious about AI readiness? See: https://www.theregister.com/2025/10/15/ciscoaireadiness_index/
- On how hyperscalers boosted Cisco's AI sales: https://www.theregister.com/2025/08/14/ciscoq42025/
- And the scrutiny over firewall issues: https://www.theregister.com/2025/10/16/ciscosenatescrutiny/
What do you think? Does empowering AI with 30 years of cyber history sound like the future of security, or do you worry about over-reliance on algorithms that could be fooled? Could this open-source approach democratize protection, or just give big players like Cisco an edge? Share your thoughts in the comments—do you agree, disagree, or have a counterpoint to add?
